audit
Audit current renders and stress coverage, repair real defects and stop under the bound refinement policy.
Implement an already chosen Figma target faithfully while verifying unrepresented behavior, states and responsive gaps.
$ npx -y skills add 3x-haust/oh-my-design --skill omd-figma --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/omd-figmaContext preview
The summary Claude sees to decide when to auto-load this skill.
Implement an already chosen Figma target faithfully while verifying unrepresented behavior, states and responsive gaps.
name: omd-figma description: Implement an already chosen Figma target faithfully while verifying unrepresented behavior, states and responsive gaps.
The supplied frames already choose their represented visual properties. Do not invent another concept tournament or silently correct the user's design. Frame the full requested surfaces, behavior/accessibility and missing responsive/state coverage; absence in Figma is a gap, not approval.
Run omd doctor. Check only whether FIGMA_TOKEN is configured, never print it or read a secret into output. If missing, request private local configuration of a read-only File content token; do not invent a workaround. Use `omd figma pull <figma-url>` and `omd figma system`, inspect the actual snapshot, component variants and responsive pairing. Reuse only needed components, not a speculative library.
Hand implements within the current route. For each supplied frame/view run `omd figma diff <frame-id> <rendered-page> --json`; repair observed mismatch and re-diff. Missing paired viewports require truthful interpolation and real render checks, not a claimed Figma match. Finite attempts or a fidelity exception do not turn a failed diff into PASS. The same unresolved defect after three failed repairs needs escalation.
Keep snapshot/attribution separate from external reference assembly. If independent research was requested, use its own current browse/analysis contract. Report original-design advisories without silently changing target properties; implementation defects remain repairs. Task/access/safety obligations still apply.
Finish with current native measured/rendered evidence, independent review and guard completion under the applicable host contract. Report per-frame/view fidelity, responsive gaps, attribution and exact limitations. Design-only requests end at their handoff and make no implemented-behavior claim.
AI that designs like a human. Give OMD one request. It researches, compares real material, chooses a direction, builds, renders, and repairs internally.
Audit current renders and stress coverage, repair real defects and stop under the bound refinement policy.
Interpret recurring measured findings and honest trends without turning taste into a skill score.
Dispatch the existing candidate-generation stage from minimal tokens to one authentic direction choice.
Review actual current design evidence and rank observable problems without implementing changes.
Expand the selected sample into the complete requested surface, state and viewport scope.